Automated Diagnosis of Drone Crashes
Drones are becoming increasingly popular in different applications however at the same time, the number of drone incidents is increasing. One way to improve the safety of drones is to look at the vast amount of data produced by them and automatically learn the causes of the incidents. This way the next versions of the drone will be designed safer against similar problems. Although data of drones and data mining techniques are available, still a methodology is needed to transform the data mining outputs into useful diagnosis results. The focus of this practical course is on using data mining methods to deduce the causal relationship between components or events in the context of cyber-physical systems in general and drones in particular.
The topics to be covered in this practical course include (but not limited to):
- CPS components and architecture
- Rule-based diagnosis
- Data-driven diagnosis
- Causality concepts and definitions
- Time series data mining
- Diagnosis of real-world drone crashes
The course is divided to four phases. In the first phase general information about the drone data and examples of shortcomings of classic rule-based diagnosis will be provided. In the second phase, a causality concept (improved prediction) will be introduced. Moreover, regression methods will be used to identify the causal relationships. The identified causal relationships will be represented in a causal model. Till here the general framework will be established. In the third phase, each group will investigate other definitions of causality (e.g. counterfactual reasoning) and other data mining methods (e.g. classification and clustering) to infer the causal model from the drone data. In the fourth phase the discovered causal model will be used for reasoning about the root cause of the incident. (More details can be found in the slides)
 Pearl, J. (2009). Causality. Cambridge university press.
 Larose, D. T.; Larose, C. D. (2014). Discovering knowledge in data: an introduction to data mining. John Wiley and Sons.
 Esling, P.; Agon, C. (2012). Time-series data mining. ACM Computing Surveys, 45(1), 12.